好調になってきた。打率の行方は上昇しかない!!
spamtrainer のログを見ると
# cat spamtrainer.log
Apr 14 01:01:55 – spamtrainer Version 1.6.1
0.000 0 191 0 non-token data: ham
0.000 0 929 0 non-token data: spam
Apr 15 01:00:13 – spamtrainer Version 1.6.1
0.000 0 203 0 non-token data: ham
0.000 0 929 0 non-token data: spam
Apr 16 01:00:57 – spamtrainer Version 1.6.1
0.000 0 213 0 non-token data: ham
0.000 0 937 0 non-token data: spam
#
とある。読み方がわからん。ま、午前1時になにかやっていることは確実だ。
SpamAssassin Training Error
によると、
Apparently everyone agrees that Apple’s SA implementation is either messed up or deliberately hobbled.
I ended up creating the symbolic link as described in several places:
sudo -s
cd /var/amavis
mv .spamassassin .spamassassin.old
su clamav
ln -s /var/clamav/.spamassassin /var/amavis/.spamassassin
…then fixing the /etc/mail/spamassassin/local.cf file:
change ‘auto_learn’ to ‘bayes_auto_learn’
change ‘safe_reporting’ to ‘report_safe’
comment out ‘use_terse_report 0’
comment out ‘subject_tag’ and ‘rewrite_subject’
(note that it is “report_safe” and not “safe_report” as indicated in some places)
…then setting up the junkmail and notjunkmail accounts as Apple recommends
…then using spamtrainer
Now, after scanning 200+ spams and 200+ non-spams (which takes quite a long time with spamtrainer), Bayesian scanning appears to be working. I can tell because the X-Spam-Status tag in my mail headers now contains a BAYES tag:
X-Spam-Status: Yes, hits=5.33 tagged_above=-999 required=5 tests=BAYES_50, HELO_DYNAMIC_IPADDR2, HTML_40_50, HTML_MESSAGE, MIME_HTML_ONLY
That was NOT trivial (considering that this is supposed to already be a feature of Tiger). The discussion in the afp548 thread starts on 10.4.2. This misconfiguration (deliberate hobbling?) persists in 10.4.6. wtf?
だそうで、結局、いままで学習されていないかったのではないかと思う今日この頃、皆さん、いかがお過ごしですか?(きっこ風)